Recurrent fuzzy logic in speech recognition

Abstract

In this paper, a novel method is presented to combine neural nets with fuzzy logic. The combined technology is based on modified Neufuz (‘71, ‘21, ‘31) using recurrent neural networks. The recurrent information of neural net is directly mapped to a new type of fuzzy logic, called “recurrent” fuzzy logic. Recurrency preserves temporal information and yields superior performance for context dependent applications like handwriting, pattem and speech recognition. It also reduces the convergence time to learn fuzzy logic rules and membership functions. We have used recurrent fuzzy logic approach to solve several problems associated with speech recognition. Simulations show good improvements in accuracx speed of learning and speaker variability for isolated word recognition.